2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@redhat.com>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
12 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
13 * added netlink_proto_exit
14 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
15 * use nlk_sk, as sk->protinfo is on a diet 8)
16 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
17 * - inc module use count of module that owns
18 * the kernel socket in case userspace opens
19 * socket of same protocol
20 * - remove all module support, since netlink is
21 * mandatory if CONFIG_NET=y these days
24 #include <linux/module.h>
26 #include <linux/capability.h>
27 #include <linux/kernel.h>
28 #include <linux/init.h>
29 #include <linux/signal.h>
30 #include <linux/sched.h>
31 #include <linux/errno.h>
32 #include <linux/string.h>
33 #include <linux/stat.h>
34 #include <linux/socket.h>
36 #include <linux/fcntl.h>
37 #include <linux/termios.h>
38 #include <linux/sockios.h>
39 #include <linux/net.h>
41 #include <linux/slab.h>
42 #include <asm/uaccess.h>
43 #include <linux/skbuff.h>
44 #include <linux/netdevice.h>
45 #include <linux/rtnetlink.h>
46 #include <linux/proc_fs.h>
47 #include <linux/seq_file.h>
48 #include <linux/smp_lock.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/selinux.h>
59 #include <linux/vs_context.h>
60 #include <linux/vs_network.h>
61 #include <linux/vs_limit.h>
65 #include <net/netlink.h>
67 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
70 /* struct sock has to be the first member of netlink_sock */
78 unsigned long *groups;
80 wait_queue_head_t wait;
81 struct netlink_callback *cb;
83 void (*data_ready)(struct sock *sk, int bytes);
84 struct module *module;
87 #define NETLINK_KERNEL_SOCKET 0x1
88 #define NETLINK_RECV_PKTINFO 0x2
90 static inline struct netlink_sock *nlk_sk(struct sock *sk)
92 return (struct netlink_sock *)sk;
96 struct hlist_head *table;
97 unsigned long rehash_time;
102 unsigned int entries;
103 unsigned int max_shift;
108 struct netlink_table {
109 struct nl_pid_hash hash;
110 struct hlist_head mc_list;
111 unsigned long *listeners;
112 unsigned int nl_nonroot;
114 struct module *module;
118 static struct netlink_table *nl_table;
120 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
122 static int netlink_dump(struct sock *sk);
123 static void netlink_destroy_callback(struct netlink_callback *cb);
125 static DEFINE_RWLOCK(nl_table_lock);
126 static atomic_t nl_table_users = ATOMIC_INIT(0);
128 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
130 static u32 netlink_group_mask(u32 group)
132 return group ? 1 << (group - 1) : 0;
135 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
137 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
140 static void netlink_sock_destruct(struct sock *sk)
142 skb_queue_purge(&sk->sk_receive_queue);
144 if (!sock_flag(sk, SOCK_DEAD)) {
145 printk("Freeing alive netlink socket %p\n", sk);
148 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
149 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
150 BUG_TRAP(!nlk_sk(sk)->cb);
151 BUG_TRAP(!nlk_sk(sk)->groups);
154 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
155 * Look, when several writers sleep and reader wakes them up, all but one
156 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
157 * this, _but_ remember, it adds useless work on UP machines.
160 static void netlink_table_grab(void)
162 write_lock_irq(&nl_table_lock);
164 if (atomic_read(&nl_table_users)) {
165 DECLARE_WAITQUEUE(wait, current);
167 add_wait_queue_exclusive(&nl_table_wait, &wait);
169 set_current_state(TASK_UNINTERRUPTIBLE);
170 if (atomic_read(&nl_table_users) == 0)
172 write_unlock_irq(&nl_table_lock);
174 write_lock_irq(&nl_table_lock);
177 __set_current_state(TASK_RUNNING);
178 remove_wait_queue(&nl_table_wait, &wait);
182 static __inline__ void netlink_table_ungrab(void)
184 write_unlock_irq(&nl_table_lock);
185 wake_up(&nl_table_wait);
188 static __inline__ void
189 netlink_lock_table(void)
191 /* read_lock() synchronizes us to netlink_table_grab */
193 read_lock(&nl_table_lock);
194 atomic_inc(&nl_table_users);
195 read_unlock(&nl_table_lock);
198 static __inline__ void
199 netlink_unlock_table(void)
201 if (atomic_dec_and_test(&nl_table_users))
202 wake_up(&nl_table_wait);
205 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
207 struct nl_pid_hash *hash = &nl_table[protocol].hash;
208 struct hlist_head *head;
210 struct hlist_node *node;
212 read_lock(&nl_table_lock);
213 head = nl_pid_hashfn(hash, pid);
214 sk_for_each(sk, node, head) {
215 if (nlk_sk(sk)->pid == pid) {
222 read_unlock(&nl_table_lock);
226 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
228 if (size <= PAGE_SIZE)
229 return kmalloc(size, GFP_ATOMIC);
231 return (struct hlist_head *)
232 __get_free_pages(GFP_ATOMIC, get_order(size));
235 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
237 if (size <= PAGE_SIZE)
240 free_pages((unsigned long)table, get_order(size));
243 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
245 unsigned int omask, mask, shift;
247 struct hlist_head *otable, *table;
250 omask = mask = hash->mask;
251 osize = size = (mask + 1) * sizeof(*table);
255 if (++shift > hash->max_shift)
261 table = nl_pid_hash_alloc(size);
265 memset(table, 0, size);
266 otable = hash->table;
270 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
272 for (i = 0; i <= omask; i++) {
274 struct hlist_node *node, *tmp;
276 sk_for_each_safe(sk, node, tmp, &otable[i])
277 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
280 nl_pid_hash_free(otable, osize);
281 hash->rehash_time = jiffies + 10 * 60 * HZ;
285 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
287 int avg = hash->entries >> hash->shift;
289 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
292 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
293 nl_pid_hash_rehash(hash, 0);
300 static const struct proto_ops netlink_ops;
303 netlink_update_listeners(struct sock *sk)
305 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
306 struct hlist_node *node;
310 for (i = 0; i < NLGRPSZ(tbl->groups)/sizeof(unsigned long); i++) {
312 sk_for_each_bound(sk, node, &tbl->mc_list)
313 mask |= nlk_sk(sk)->groups[i];
314 tbl->listeners[i] = mask;
316 /* this function is only called with the netlink table "grabbed", which
317 * makes sure updates are visible before bind or setsockopt return. */
320 static int netlink_insert(struct sock *sk, u32 pid)
322 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
323 struct hlist_head *head;
324 int err = -EADDRINUSE;
326 struct hlist_node *node;
329 netlink_table_grab();
330 head = nl_pid_hashfn(hash, pid);
332 sk_for_each(osk, node, head) {
333 if (nlk_sk(osk)->pid == pid)
345 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
348 if (len && nl_pid_hash_dilute(hash, len))
349 head = nl_pid_hashfn(hash, pid);
351 nlk_sk(sk)->pid = pid;
352 sk_add_node(sk, head);
356 netlink_table_ungrab();
360 static void netlink_remove(struct sock *sk)
362 netlink_table_grab();
363 if (sk_del_node_init(sk))
364 nl_table[sk->sk_protocol].hash.entries--;
365 if (nlk_sk(sk)->subscriptions)
366 __sk_del_bind_node(sk);
367 netlink_table_ungrab();
370 static struct proto netlink_proto = {
372 .owner = THIS_MODULE,
373 .obj_size = sizeof(struct netlink_sock),
376 static int __netlink_create(struct socket *sock, int protocol)
379 struct netlink_sock *nlk;
381 sock->ops = &netlink_ops;
383 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
387 sock_init_data(sock, sk);
390 spin_lock_init(&nlk->cb_lock);
391 init_waitqueue_head(&nlk->wait);
393 sk->sk_destruct = netlink_sock_destruct;
394 sk->sk_protocol = protocol;
398 static int netlink_create(struct socket *sock, int protocol)
400 struct module *module = NULL;
401 struct netlink_sock *nlk;
405 sock->state = SS_UNCONNECTED;
407 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
408 return -ESOCKTNOSUPPORT;
410 if (protocol<0 || protocol >= MAX_LINKS)
411 return -EPROTONOSUPPORT;
413 netlink_lock_table();
415 if (!nl_table[protocol].registered) {
416 netlink_unlock_table();
417 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
418 netlink_lock_table();
421 if (nl_table[protocol].registered &&
422 try_module_get(nl_table[protocol].module))
423 module = nl_table[protocol].module;
424 groups = nl_table[protocol].groups;
425 netlink_unlock_table();
427 if ((err = __netlink_create(sock, protocol)) < 0)
430 nlk = nlk_sk(sock->sk);
431 nlk->module = module;
440 static int netlink_release(struct socket *sock)
442 struct sock *sk = sock->sk;
443 struct netlink_sock *nlk;
451 spin_lock(&nlk->cb_lock);
454 nlk->cb->done(nlk->cb);
455 netlink_destroy_callback(nlk->cb);
458 spin_unlock(&nlk->cb_lock);
460 /* OK. Socket is unlinked, and, therefore,
461 no new packets will arrive */
465 wake_up_interruptible_all(&nlk->wait);
467 skb_queue_purge(&sk->sk_write_queue);
469 if (nlk->pid && !nlk->subscriptions) {
470 struct netlink_notify n = {
471 .protocol = sk->sk_protocol,
474 atomic_notifier_call_chain(&netlink_chain,
475 NETLINK_URELEASE, &n);
478 module_put(nlk->module);
480 netlink_table_grab();
481 if (nlk->flags & NETLINK_KERNEL_SOCKET) {
482 kfree(nl_table[sk->sk_protocol].listeners);
483 nl_table[sk->sk_protocol].module = NULL;
484 nl_table[sk->sk_protocol].registered = 0;
485 } else if (nlk->subscriptions)
486 netlink_update_listeners(sk);
487 netlink_table_ungrab();
496 static int netlink_autobind(struct socket *sock)
498 struct sock *sk = sock->sk;
499 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
500 struct hlist_head *head;
502 struct hlist_node *node;
503 s32 pid = current->tgid;
505 static s32 rover = -4097;
509 netlink_table_grab();
510 head = nl_pid_hashfn(hash, pid);
511 sk_for_each(osk, node, head) {
512 if (nlk_sk(osk)->pid == pid) {
513 /* Bind collision, search negative pid values. */
517 netlink_table_ungrab();
521 netlink_table_ungrab();
523 err = netlink_insert(sk, pid);
524 if (err == -EADDRINUSE)
527 /* If 2 threads race to autobind, that is fine. */
534 static inline int netlink_capable(struct socket *sock, unsigned int flag)
536 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
537 capable(CAP_NET_ADMIN);
541 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
543 struct netlink_sock *nlk = nlk_sk(sk);
545 if (nlk->subscriptions && !subscriptions)
546 __sk_del_bind_node(sk);
547 else if (!nlk->subscriptions && subscriptions)
548 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
549 nlk->subscriptions = subscriptions;
552 static int netlink_alloc_groups(struct sock *sk)
554 struct netlink_sock *nlk = nlk_sk(sk);
558 netlink_lock_table();
559 groups = nl_table[sk->sk_protocol].groups;
560 if (!nl_table[sk->sk_protocol].registered)
562 netlink_unlock_table();
567 nlk->groups = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
568 if (nlk->groups == NULL)
570 nlk->ngroups = groups;
574 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
576 struct sock *sk = sock->sk;
577 struct netlink_sock *nlk = nlk_sk(sk);
578 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
581 if (nladdr->nl_family != AF_NETLINK)
584 /* Only superuser is allowed to listen multicasts */
585 if (nladdr->nl_groups) {
586 if (!netlink_capable(sock, NL_NONROOT_RECV))
588 if (nlk->groups == NULL) {
589 err = netlink_alloc_groups(sk);
596 if (nladdr->nl_pid != nlk->pid)
599 err = nladdr->nl_pid ?
600 netlink_insert(sk, nladdr->nl_pid) :
601 netlink_autobind(sock);
606 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
609 netlink_table_grab();
610 netlink_update_subscriptions(sk, nlk->subscriptions +
611 hweight32(nladdr->nl_groups) -
612 hweight32(nlk->groups[0]));
613 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
614 netlink_update_listeners(sk);
615 netlink_table_ungrab();
620 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
624 struct sock *sk = sock->sk;
625 struct netlink_sock *nlk = nlk_sk(sk);
626 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
628 if (addr->sa_family == AF_UNSPEC) {
629 sk->sk_state = NETLINK_UNCONNECTED;
634 if (addr->sa_family != AF_NETLINK)
637 /* Only superuser is allowed to send multicasts */
638 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
642 err = netlink_autobind(sock);
645 sk->sk_state = NETLINK_CONNECTED;
646 nlk->dst_pid = nladdr->nl_pid;
647 nlk->dst_group = ffs(nladdr->nl_groups);
653 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
655 struct sock *sk = sock->sk;
656 struct netlink_sock *nlk = nlk_sk(sk);
657 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
659 nladdr->nl_family = AF_NETLINK;
661 *addr_len = sizeof(*nladdr);
664 nladdr->nl_pid = nlk->dst_pid;
665 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
667 nladdr->nl_pid = nlk->pid;
668 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
673 static void netlink_overrun(struct sock *sk)
675 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
676 sk->sk_err = ENOBUFS;
677 sk->sk_error_report(sk);
681 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
683 int protocol = ssk->sk_protocol;
685 struct netlink_sock *nlk;
687 sock = netlink_lookup(protocol, pid);
689 return ERR_PTR(-ECONNREFUSED);
691 /* Don't bother queuing skb if kernel socket has no input function */
693 if ((nlk->pid == 0 && !nlk->data_ready) ||
694 (sock->sk_state == NETLINK_CONNECTED &&
695 nlk->dst_pid != nlk_sk(ssk)->pid)) {
697 return ERR_PTR(-ECONNREFUSED);
702 struct sock *netlink_getsockbyfilp(struct file *filp)
704 struct inode *inode = filp->f_path.dentry->d_inode;
707 if (!S_ISSOCK(inode->i_mode))
708 return ERR_PTR(-ENOTSOCK);
710 sock = SOCKET_I(inode)->sk;
711 if (sock->sk_family != AF_NETLINK)
712 return ERR_PTR(-EINVAL);
719 * Attach a skb to a netlink socket.
720 * The caller must hold a reference to the destination socket. On error, the
721 * reference is dropped. The skb is not send to the destination, just all
722 * all error checks are performed and memory in the queue is reserved.
724 * < 0: error. skb freed, reference to sock dropped.
726 * 1: repeat lookup - reference dropped while waiting for socket memory.
728 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
729 long timeo, struct sock *ssk)
731 struct netlink_sock *nlk;
735 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
736 test_bit(0, &nlk->state)) {
737 DECLARE_WAITQUEUE(wait, current);
739 if (!ssk || nlk_sk(ssk)->pid == 0)
746 __set_current_state(TASK_INTERRUPTIBLE);
747 add_wait_queue(&nlk->wait, &wait);
749 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
750 test_bit(0, &nlk->state)) &&
751 !sock_flag(sk, SOCK_DEAD))
752 timeo = schedule_timeout(timeo);
754 __set_current_state(TASK_RUNNING);
755 remove_wait_queue(&nlk->wait, &wait);
758 if (signal_pending(current)) {
760 return sock_intr_errno(timeo);
764 skb_set_owner_r(skb, sk);
768 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
772 skb_queue_tail(&sk->sk_receive_queue, skb);
773 sk->sk_data_ready(sk, len);
778 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
784 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
791 delta = skb->end - skb->tail;
792 if (delta * 2 < skb->truesize)
795 if (skb_shared(skb)) {
796 struct sk_buff *nskb = skb_clone(skb, allocation);
803 if (!pskb_expand_head(skb, 0, -delta, allocation))
804 skb->truesize -= delta;
809 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
815 skb = netlink_trim(skb, gfp_any());
817 timeo = sock_sndtimeo(ssk, nonblock);
819 sk = netlink_getsockbypid(ssk, pid);
824 err = netlink_attachskb(sk, skb, nonblock, timeo, ssk);
830 return netlink_sendskb(sk, skb, ssk->sk_protocol);
833 int netlink_has_listeners(struct sock *sk, unsigned int group)
837 BUG_ON(!(nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET));
838 if (group - 1 < nl_table[sk->sk_protocol].groups)
839 res = test_bit(group - 1, nl_table[sk->sk_protocol].listeners);
842 EXPORT_SYMBOL_GPL(netlink_has_listeners);
844 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
846 struct netlink_sock *nlk = nlk_sk(sk);
848 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
849 !test_bit(0, &nlk->state)) {
850 skb_set_owner_r(skb, sk);
851 skb_queue_tail(&sk->sk_receive_queue, skb);
852 sk->sk_data_ready(sk, skb->len);
853 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
858 struct netlink_broadcast_data {
859 struct sock *exclude_sk;
866 struct sk_buff *skb, *skb2;
869 static inline int do_one_broadcast(struct sock *sk,
870 struct netlink_broadcast_data *p)
872 struct netlink_sock *nlk = nlk_sk(sk);
875 if (p->exclude_sk == sk)
878 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
879 !test_bit(p->group - 1, nlk->groups))
888 if (p->skb2 == NULL) {
889 if (skb_shared(p->skb)) {
890 p->skb2 = skb_clone(p->skb, p->allocation);
892 p->skb2 = skb_get(p->skb);
894 * skb ownership may have been set when
895 * delivered to a previous socket.
900 if (p->skb2 == NULL) {
902 /* Clone failed. Notify ALL listeners. */
904 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
917 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
918 u32 group, gfp_t allocation)
920 struct netlink_broadcast_data info;
921 struct hlist_node *node;
924 skb = netlink_trim(skb, allocation);
926 info.exclude_sk = ssk;
932 info.allocation = allocation;
936 /* While we sleep in clone, do not allow to change socket list */
938 netlink_lock_table();
940 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
941 do_one_broadcast(sk, &info);
945 netlink_unlock_table();
948 kfree_skb(info.skb2);
950 if (info.delivered) {
951 if (info.congested && (allocation & __GFP_WAIT))
960 struct netlink_set_err_data {
961 struct sock *exclude_sk;
967 static inline int do_one_set_err(struct sock *sk,
968 struct netlink_set_err_data *p)
970 struct netlink_sock *nlk = nlk_sk(sk);
972 if (sk == p->exclude_sk)
975 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
976 !test_bit(p->group - 1, nlk->groups))
979 sk->sk_err = p->code;
980 sk->sk_error_report(sk);
985 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
987 struct netlink_set_err_data info;
988 struct hlist_node *node;
991 info.exclude_sk = ssk;
996 read_lock(&nl_table_lock);
998 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
999 do_one_set_err(sk, &info);
1001 read_unlock(&nl_table_lock);
1004 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1005 char __user *optval, int optlen)
1007 struct sock *sk = sock->sk;
1008 struct netlink_sock *nlk = nlk_sk(sk);
1011 if (level != SOL_NETLINK)
1012 return -ENOPROTOOPT;
1014 if (optlen >= sizeof(int) &&
1015 get_user(val, (int __user *)optval))
1019 case NETLINK_PKTINFO:
1021 nlk->flags |= NETLINK_RECV_PKTINFO;
1023 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1026 case NETLINK_ADD_MEMBERSHIP:
1027 case NETLINK_DROP_MEMBERSHIP: {
1028 unsigned int subscriptions;
1029 int old, new = optname == NETLINK_ADD_MEMBERSHIP ? 1 : 0;
1031 if (!netlink_capable(sock, NL_NONROOT_RECV))
1033 if (nlk->groups == NULL) {
1034 err = netlink_alloc_groups(sk);
1038 if (!val || val - 1 >= nlk->ngroups)
1040 netlink_table_grab();
1041 old = test_bit(val - 1, nlk->groups);
1042 subscriptions = nlk->subscriptions - old + new;
1044 __set_bit(val - 1, nlk->groups);
1046 __clear_bit(val - 1, nlk->groups);
1047 netlink_update_subscriptions(sk, subscriptions);
1048 netlink_update_listeners(sk);
1049 netlink_table_ungrab();
1059 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1060 char __user *optval, int __user *optlen)
1062 struct sock *sk = sock->sk;
1063 struct netlink_sock *nlk = nlk_sk(sk);
1066 if (level != SOL_NETLINK)
1067 return -ENOPROTOOPT;
1069 if (get_user(len, optlen))
1075 case NETLINK_PKTINFO:
1076 if (len < sizeof(int))
1079 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1080 if (put_user(len, optlen) ||
1081 put_user(val, optval))
1091 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1093 struct nl_pktinfo info;
1095 info.group = NETLINK_CB(skb).dst_group;
1096 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1099 static inline void netlink_rcv_wake(struct sock *sk)
1101 struct netlink_sock *nlk = nlk_sk(sk);
1103 if (skb_queue_empty(&sk->sk_receive_queue))
1104 clear_bit(0, &nlk->state);
1105 if (!test_bit(0, &nlk->state))
1106 wake_up_interruptible(&nlk->wait);
1109 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1110 struct msghdr *msg, size_t len)
1112 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1113 struct sock *sk = sock->sk;
1114 struct netlink_sock *nlk = nlk_sk(sk);
1115 struct sockaddr_nl *addr=msg->msg_name;
1118 struct sk_buff *skb;
1120 struct scm_cookie scm;
1122 if (msg->msg_flags&MSG_OOB)
1125 if (NULL == siocb->scm)
1127 err = scm_send(sock, msg, siocb->scm);
1131 if (msg->msg_namelen) {
1132 if (addr->nl_family != AF_NETLINK)
1134 dst_pid = addr->nl_pid;
1135 dst_group = ffs(addr->nl_groups);
1136 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1139 dst_pid = nlk->dst_pid;
1140 dst_group = nlk->dst_group;
1144 err = netlink_autobind(sock);
1150 if (len > sk->sk_sndbuf - 32)
1153 skb = alloc_skb(len, GFP_KERNEL);
1157 NETLINK_CB(skb).pid = nlk->pid;
1158 NETLINK_CB(skb).dst_group = dst_group;
1159 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1160 selinux_get_task_sid(current, &(NETLINK_CB(skb).sid));
1161 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1163 /* What can I do? Netlink is asynchronous, so that
1164 we will have to save current capabilities to
1165 check them, when this message will be delivered
1166 to corresponding kernel module. --ANK (980802)
1170 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1175 err = security_netlink_send(sk, skb);
1182 atomic_inc(&skb->users);
1183 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1185 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1191 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1192 struct msghdr *msg, size_t len,
1195 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1196 struct scm_cookie scm;
1197 struct sock *sk = sock->sk;
1198 struct netlink_sock *nlk = nlk_sk(sk);
1199 int noblock = flags&MSG_DONTWAIT;
1201 struct sk_buff *skb;
1209 skb = skb_recv_datagram(sk,flags,noblock,&err);
1213 msg->msg_namelen = 0;
1217 msg->msg_flags |= MSG_TRUNC;
1221 skb->h.raw = skb->data;
1222 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1224 if (msg->msg_name) {
1225 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1226 addr->nl_family = AF_NETLINK;
1228 addr->nl_pid = NETLINK_CB(skb).pid;
1229 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1230 msg->msg_namelen = sizeof(*addr);
1233 if (nlk->flags & NETLINK_RECV_PKTINFO)
1234 netlink_cmsg_recv_pktinfo(msg, skb);
1236 if (NULL == siocb->scm) {
1237 memset(&scm, 0, sizeof(scm));
1240 siocb->scm->creds = *NETLINK_CREDS(skb);
1241 skb_free_datagram(sk, skb);
1243 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1246 scm_recv(sock, msg, siocb->scm, flags);
1249 netlink_rcv_wake(sk);
1250 return err ? : copied;
1253 static void netlink_data_ready(struct sock *sk, int len)
1255 struct netlink_sock *nlk = nlk_sk(sk);
1257 if (nlk->data_ready)
1258 nlk->data_ready(sk, len);
1259 netlink_rcv_wake(sk);
1263 * We export these functions to other modules. They provide a
1264 * complete set of kernel non-blocking support for message
1269 netlink_kernel_create(int unit, unsigned int groups,
1270 void (*input)(struct sock *sk, int len),
1271 struct module *module)
1273 struct socket *sock;
1275 struct netlink_sock *nlk;
1276 unsigned long *listeners = NULL;
1280 if (unit<0 || unit>=MAX_LINKS)
1283 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1286 if (__netlink_create(sock, unit) < 0)
1287 goto out_sock_release;
1292 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1294 goto out_sock_release;
1297 sk->sk_data_ready = netlink_data_ready;
1299 nlk_sk(sk)->data_ready = input;
1301 if (netlink_insert(sk, 0))
1302 goto out_sock_release;
1305 nlk->flags |= NETLINK_KERNEL_SOCKET;
1307 netlink_table_grab();
1308 nl_table[unit].groups = groups;
1309 nl_table[unit].listeners = listeners;
1310 nl_table[unit].module = module;
1311 nl_table[unit].registered = 1;
1312 netlink_table_ungrab();
1322 void netlink_set_nonroot(int protocol, unsigned int flags)
1324 if ((unsigned int)protocol < MAX_LINKS)
1325 nl_table[protocol].nl_nonroot = flags;
1328 static void netlink_destroy_callback(struct netlink_callback *cb)
1336 * It looks a bit ugly.
1337 * It would be better to create kernel thread.
1340 static int netlink_dump(struct sock *sk)
1342 struct netlink_sock *nlk = nlk_sk(sk);
1343 struct netlink_callback *cb;
1344 struct sk_buff *skb;
1345 struct nlmsghdr *nlh;
1346 int len, err = -ENOBUFS;
1348 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1352 spin_lock(&nlk->cb_lock);
1360 len = cb->dump(skb, cb);
1363 spin_unlock(&nlk->cb_lock);
1364 skb_queue_tail(&sk->sk_receive_queue, skb);
1365 sk->sk_data_ready(sk, len);
1369 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1373 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1375 skb_queue_tail(&sk->sk_receive_queue, skb);
1376 sk->sk_data_ready(sk, skb->len);
1381 spin_unlock(&nlk->cb_lock);
1383 netlink_destroy_callback(cb);
1387 spin_unlock(&nlk->cb_lock);
1393 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1394 struct nlmsghdr *nlh,
1395 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1396 int (*done)(struct netlink_callback*))
1398 struct netlink_callback *cb;
1400 struct netlink_sock *nlk;
1402 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1409 atomic_inc(&skb->users);
1412 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1414 netlink_destroy_callback(cb);
1415 return -ECONNREFUSED;
1418 /* A dump is in progress... */
1419 spin_lock(&nlk->cb_lock);
1421 spin_unlock(&nlk->cb_lock);
1422 netlink_destroy_callback(cb);
1427 spin_unlock(&nlk->cb_lock);
1434 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1436 struct sk_buff *skb;
1437 struct nlmsghdr *rep;
1438 struct nlmsgerr *errmsg;
1439 size_t payload = sizeof(*errmsg);
1441 /* error messages get the original request appened */
1443 payload += nlmsg_len(nlh);
1445 skb = nlmsg_new(payload, GFP_KERNEL);
1449 sk = netlink_lookup(in_skb->sk->sk_protocol,
1450 NETLINK_CB(in_skb).pid);
1452 sk->sk_err = ENOBUFS;
1453 sk->sk_error_report(sk);
1459 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1460 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1461 errmsg = nlmsg_data(rep);
1462 errmsg->error = err;
1463 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1464 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1467 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1468 struct nlmsghdr *, int *))
1470 struct nlmsghdr *nlh;
1473 while (skb->len >= nlmsg_total_size(0)) {
1474 nlh = (struct nlmsghdr *) skb->data;
1476 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1479 if (cb(skb, nlh, &err) < 0) {
1480 /* Not an error, but we have to interrupt processing
1481 * here. Note: that in this case we do not pull
1482 * message from skb, it will be processed later.
1486 netlink_ack(skb, nlh, err);
1487 } else if (nlh->nlmsg_flags & NLM_F_ACK)
1488 netlink_ack(skb, nlh, 0);
1490 netlink_queue_skip(nlh, skb);
1497 * nelink_run_queue - Process netlink receive queue.
1498 * @sk: Netlink socket containing the queue
1499 * @qlen: Place to store queue length upon entry
1500 * @cb: Callback function invoked for each netlink message found
1502 * Processes as much as there was in the queue upon entry and invokes
1503 * a callback function for each netlink message found. The callback
1504 * function may refuse a message by returning a negative error code
1505 * but setting the error pointer to 0 in which case this function
1506 * returns with a qlen != 0.
1508 * qlen must be initialized to 0 before the initial entry, afterwards
1509 * the function may be called repeatedly until qlen reaches 0.
1511 void netlink_run_queue(struct sock *sk, unsigned int *qlen,
1512 int (*cb)(struct sk_buff *, struct nlmsghdr *, int *))
1514 struct sk_buff *skb;
1516 if (!*qlen || *qlen > skb_queue_len(&sk->sk_receive_queue))
1517 *qlen = skb_queue_len(&sk->sk_receive_queue);
1519 for (; *qlen; (*qlen)--) {
1520 skb = skb_dequeue(&sk->sk_receive_queue);
1521 if (netlink_rcv_skb(skb, cb)) {
1523 skb_queue_head(&sk->sk_receive_queue, skb);
1536 * netlink_queue_skip - Skip netlink message while processing queue.
1537 * @nlh: Netlink message to be skipped
1538 * @skb: Socket buffer containing the netlink messages.
1540 * Pulls the given netlink message off the socket buffer so the next
1541 * call to netlink_queue_run() will not reconsider the message.
1543 void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb)
1545 int msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1547 if (msglen > skb->len)
1550 skb_pull(skb, msglen);
1554 * nlmsg_notify - send a notification netlink message
1555 * @sk: netlink socket to use
1556 * @skb: notification message
1557 * @pid: destination netlink pid for reports or 0
1558 * @group: destination multicast group or 0
1559 * @report: 1 to report back, 0 to disable
1560 * @flags: allocation flags
1562 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1563 unsigned int group, int report, gfp_t flags)
1568 int exclude_pid = 0;
1571 atomic_inc(&skb->users);
1575 /* errors reported via destination sk->sk_err */
1576 nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1580 err = nlmsg_unicast(sk, skb, pid);
1585 #ifdef CONFIG_PROC_FS
1586 struct nl_seq_iter {
1591 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1593 struct nl_seq_iter *iter = seq->private;
1596 struct hlist_node *node;
1599 for (i=0; i<MAX_LINKS; i++) {
1600 struct nl_pid_hash *hash = &nl_table[i].hash;
1602 for (j = 0; j <= hash->mask; j++) {
1603 sk_for_each(s, node, &hash->table[j]) {
1616 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1618 read_lock(&nl_table_lock);
1619 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1622 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1625 struct nl_seq_iter *iter;
1630 if (v == SEQ_START_TOKEN)
1631 return netlink_seq_socket_idx(seq, 0);
1637 iter = seq->private;
1639 j = iter->hash_idx + 1;
1642 struct nl_pid_hash *hash = &nl_table[i].hash;
1644 for (; j <= hash->mask; j++) {
1645 s = sk_head(&hash->table[j]);
1654 } while (++i < MAX_LINKS);
1659 static void netlink_seq_stop(struct seq_file *seq, void *v)
1661 read_unlock(&nl_table_lock);
1665 static int netlink_seq_show(struct seq_file *seq, void *v)
1667 if (v == SEQ_START_TOKEN)
1669 "sk Eth Pid Groups "
1670 "Rmem Wmem Dump Locks\n");
1673 struct netlink_sock *nlk = nlk_sk(s);
1675 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1679 nlk->groups ? (u32)nlk->groups[0] : 0,
1680 atomic_read(&s->sk_rmem_alloc),
1681 atomic_read(&s->sk_wmem_alloc),
1683 atomic_read(&s->sk_refcnt)
1690 static struct seq_operations netlink_seq_ops = {
1691 .start = netlink_seq_start,
1692 .next = netlink_seq_next,
1693 .stop = netlink_seq_stop,
1694 .show = netlink_seq_show,
1698 static int netlink_seq_open(struct inode *inode, struct file *file)
1700 struct seq_file *seq;
1701 struct nl_seq_iter *iter;
1704 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1708 err = seq_open(file, &netlink_seq_ops);
1714 seq = file->private_data;
1715 seq->private = iter;
1719 static struct file_operations netlink_seq_fops = {
1720 .owner = THIS_MODULE,
1721 .open = netlink_seq_open,
1723 .llseek = seq_lseek,
1724 .release = seq_release_private,
1729 int netlink_register_notifier(struct notifier_block *nb)
1731 return atomic_notifier_chain_register(&netlink_chain, nb);
1734 int netlink_unregister_notifier(struct notifier_block *nb)
1736 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1739 static const struct proto_ops netlink_ops = {
1740 .family = PF_NETLINK,
1741 .owner = THIS_MODULE,
1742 .release = netlink_release,
1743 .bind = netlink_bind,
1744 .connect = netlink_connect,
1745 .socketpair = sock_no_socketpair,
1746 .accept = sock_no_accept,
1747 .getname = netlink_getname,
1748 .poll = datagram_poll,
1749 .ioctl = sock_no_ioctl,
1750 .listen = sock_no_listen,
1751 .shutdown = sock_no_shutdown,
1752 .setsockopt = netlink_setsockopt,
1753 .getsockopt = netlink_getsockopt,
1754 .sendmsg = netlink_sendmsg,
1755 .recvmsg = netlink_recvmsg,
1756 .mmap = sock_no_mmap,
1757 .sendpage = sock_no_sendpage,
1760 static struct net_proto_family netlink_family_ops = {
1761 .family = PF_NETLINK,
1762 .create = netlink_create,
1763 .owner = THIS_MODULE, /* for consistency 8) */
1766 static int __init netlink_proto_init(void)
1768 struct sk_buff *dummy_skb;
1772 int err = proto_register(&netlink_proto, 0);
1777 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1779 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1783 if (num_physpages >= (128 * 1024))
1784 max = num_physpages >> (21 - PAGE_SHIFT);
1786 max = num_physpages >> (23 - PAGE_SHIFT);
1788 order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1789 max = (1UL << order) / sizeof(struct hlist_head);
1790 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1792 for (i = 0; i < MAX_LINKS; i++) {
1793 struct nl_pid_hash *hash = &nl_table[i].hash;
1795 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1798 nl_pid_hash_free(nl_table[i].hash.table,
1799 1 * sizeof(*hash->table));
1803 memset(hash->table, 0, 1 * sizeof(*hash->table));
1804 hash->max_shift = order;
1807 hash->rehash_time = jiffies;
1810 sock_register(&netlink_family_ops);
1811 #ifdef CONFIG_PROC_FS
1812 proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1814 /* The netlink device handler may be needed early. */
1819 panic("netlink_init: Cannot allocate nl_table\n");
1822 core_initcall(netlink_proto_init);
1824 EXPORT_SYMBOL(netlink_ack);
1825 EXPORT_SYMBOL(netlink_run_queue);
1826 EXPORT_SYMBOL(netlink_queue_skip);
1827 EXPORT_SYMBOL(netlink_broadcast);
1828 EXPORT_SYMBOL(netlink_dump_start);
1829 EXPORT_SYMBOL(netlink_kernel_create);
1830 EXPORT_SYMBOL(netlink_register_notifier);
1831 EXPORT_SYMBOL(netlink_set_err);
1832 EXPORT_SYMBOL(netlink_set_nonroot);
1833 EXPORT_SYMBOL(netlink_unicast);
1834 EXPORT_SYMBOL(netlink_unregister_notifier);
1835 EXPORT_SYMBOL(nlmsg_notify);